Multi-Tissue Derived Windowing Technology Based on Statistical Features and Its Colorization Application
Jiaji Liu, Huiyan Jiang, Zexu Zhang, Guoxiu Lu · 2021
Traditional CT data processing methods often employ the inherent window to clip the data and store the image. For some DICOM-type data processing software, although the window can be adjusted dynamically, it still requires users’ the medical knowledge and familiarity in medical images. In this work, we propose a derivative windowing process method (multi-tissue derived windowing technology) for multiple tissues and construct organs based on the statistical features of CT images. Then, we design a transformation function to convert the density value of CT image into gray value. Finally, we construct the automatic colorization network using Dense U-Net as the generator of Generation Adversarial Network(GAN) model, making the multiple organs of the human body colored in a globally optimized way to highlight the color texture features of them. In this way the quality of color images can be improved significantly. In the test study, we evaluate the proposed method using many image quality evaluation methods and subjective visual evaluation on the 3Dircadb data set. The experimental results demonstrate the effectiveness of the proposed method, which indeed improves the quality of the image.